Recommendation on using cNMF for sample-level subtyping

[linlejiang]

Hi @dylkot,

Congratulations on your new publication — great work!

I’m currently following your analyze_batcheffectcorrect_BaronEtAl.ipynb tutorial to analyze my own data. I’m new to this and would really appreciate your opinion on one point:

Would you recommend using NMF for sample subtyping?

I understand that one could aggregate the cell-level usage matrix or score the top genes per GEP to obtain sample-level features, but I’d be very interested to hear your thoughts on the best approach for this kind of analysis.

Thank you!

[dylkot]

Hi @linlejiang, thanks for using the tool. I don't actually understand what you mean by sample subtyping in this case.

[linlejiang]

Hi @dylkot, thanks for the quick reply!

By sample subtyping, I meant identifying potential molecular subtypes or classifications of samples (e.g., grouping patient samples based on shared transcriptional programs).

I was wondering if it makes sense to use the GEPs from cNMF — say, by aggregating cell-level usage scores or summarizing top genes per GEP at the sample level — to explore such sample-level subclusters or disease subtypes.

Do you think this would be a reasonable application of cNMF, or is there a better approach for capturing that kind of structure across samples?